Trader consensus on Polymarket reflects a 77.5% implied probability for "No" as frontier AI models remain far from scoring 90% on the FrontierMath benchmark before 2027, with the leaderboard leader—OpenAI's GPT-5.5 Pro—achieving just 52.4% as of May 13, 2026. This ultra-challenging evaluation of research-level mathematics problems, featuring unsolved open questions, has seen steady gains from earlier sub-10% scores, driven by recent releases like GPT-5.5 and Claude Opus 4, but progress has plateaued below 55% amid scaling limits and verification hurdles. Traders anticipate no near-term breakthrough, citing historical benchmark trends where math frontiers lag reasoning gains; key catalysts include summer model drops from Anthropic or xAI, though regulatory scrutiny on AI safety may delay frontier pushes.